MedScan

Inspiration

The inspiration for this project came from the increasing role of AI in healthcare. With the growing need for quick and accurate diagnoses, we wanted to build a tool that could assist healthcare professionals by leveraging AI to analyze medical images and provide insights. Additionally, we wanted to integrate a chat feature that would allow doctors to ask AI-driven questions about medical conditions.

What We Learned

Through this project, we deepened my understanding of several key technologies:

• Flask: Used as the backbone for the web application, helping manage routes and API calls.
  
• Bootstrap, HTML, and CSS: Designed a simple yet functional interface for healthcare professionals.
  
• TorchXRayVision: Learned how to integrate a pretrained deep learning model for medical image analysis.
  
• Hugging Face Qwen 2: Implemented a large language model (LLM) to assist doctors in answering medical-related questions.
  
• Model Deployment: Gained experience in handling AI model inference in a web application.
  

How We Built It

1. Setting Up the Backend: Used Flask to create the web server and handle API requests for AI-powered image analysis and the LLM chat feature.
   
2. Integrating AI Models:
   
   • Used TorchXRayVision to process medical images and provide diagnostic predictions.
     
   • Integrated Hugging Face Qwen 2 to power the chatbot for answering medical-related queries.
     
3. Building the Frontend: Developed a responsive UI with Bootstrap, HTML, and CSS to ensure ease of use.
   
4. Connecting Everything: Implemented API calls between the frontend and backend to allow seamless interactions with the AI models.
   

Challenges Faced

• Model Inference Speed: Running AI models on a web server required optimization to ensure fast responses for both image analysis and the chatbot.
  
• Handling Large Image Files: Managing medical image uploads efficiently while maintaining performance was a key challenge.
  
• Fine-Tuning AI Responses: Ensuring that the chatbot provided relevant and accurate medical-related answers required careful prompt engineering and testing.
  

Conclusion

This project was a great learning experience, combining AI, web development, and healthcare. By leveraging pretrained AI models, we were able to create a tool that could assist doctors in both diagnosing medical images and answering medical questions. Moving forward, we aim to enhance the model’s accuracy, improve real-time performance, and explore potential regulatory compliance for wider adoption in healthcare settings.

Built With

• flask
• html/css
• python
• pytorch